Zack Edds

Zack Edds

Computer Science at the University of Illinois Urbana-Champaign. I build LLM agents, fine-tune models, ship software, and chase hard problems.

Currently back at Apple for a second summer. Previously at Argonne National Laboratory and LanzaJet.

Resume Projects
Zack Edds

What I've been working on

Terminal Agent

Fine-tuned a 2B model with QLoRA on Apple Silicon to generate bash commands from natural language. 59% to 91% accuracy, trained in 28 minutes.

Apple

SWE intern currently on Peak Power Management. Previously on iOS 26 Books & Notes: filed 28 P1 bugs and prototyped an agentic MCP workflow for on-device bug verification.

AI Agents Research

LLM agents research under Professor Lingming Zhang at UIUC. Speculative decoding benchmarks for agent swarms, context management, and agent program analysis.

NoRules.fun

LLM PvP game with LLM-as-judge and unlimited actions. DeepSeek + FLUX image gen, 49-test eval suite, A/B tested prompts from 39% to 80% pass rate.